Manifesto · v1

We are trying to shorten the distance between a problem and a thing that solves it.

This is the document we write for ourselves when we need to remember what we are doing. It is also the document we hand to anyone thinking about joining, investing in, or trusting their work to DRONA.

01Why we started

Building a custom drone is, in 2026, still an artisanal act. A first-time builder spends an average of 22 hours opening browser tabs, cross-referencing datasheets, second-guessing compatibility, and watching the same tutorials they already watched yesterday. Around 40% of those builders still, after all of that, order at least one incompatible part. They are not stupid. The problem is that drone design is a tangle of interdependent choices — frame wheelbase against propeller diameter against motor KV against battery voltage against ESC current against flight-controller UART budget — and no single tool holds the whole tangle. Reddit holds a piece. YouTube holds another. GetFPV holds a third. The knowledge is there. It is just not assembled for you, in the moment you need it.

Meanwhile the world beyond the hobby builder has the same shape. A farmer in the mountains wants to find his cows. A solar operator wants to inspect a 2 MW rooftop. An archaeologist wants to survey a reef. None of them wake up wanting a drone. They wake up wanting an answer. The drone is the shape that answer takes, and the distance from their problem to the drone is, in every case, far too long.

That distance is what we are here to shorten.

02What we are

We are an applied-research team building the conversation layer for physical systems. Our first domain is drones because drones are the fastest-moving, most catalog-friendly, and most visually rewarding corner of consumer robotics. They are also small enough that the physics still fits inside a modestly-sized engine, large enough that the economics matter, and broad enough that a single tool can span a tiny-whoop hobbyist and a 60 kg agricultural sprayer and a reef-survey ROV.

If we get drones right, we believe the same architectural pattern extends to unmanned ground vehicles, autonomous surface vessels, fixed sensor networks, and eventually to a lot of physical infrastructure that gets designed once and never looked at again. But we do not want to promise that before we have earned the first domain. So: drones, first, for a while.

03What we believe about product

Principle 01
Every recommendation is buildable.

No hallucinated part numbers, no invented KV ratings, no stack-mount patterns that do not physically exist. If the chat recommends it, the part is in our catalog with verified evidence or it is flagged as estimated. This discipline is the reason we put the physics outside the language model and expose it through typed tool calls. The AI is the interface; the catalog and the compatibility engine are the authorities.

Principle 02
Describe the job, not the airframe.

A user never has to know the difference between a deadcat and a stretched-X. They describe what they want to do. The chat picks the form factor, explains the choice in one sentence, and offers one alternative. This is the single most important UX rule in the product.

Principle 03
Watching the drone assemble in 3D is the product, not decoration.

It is the moment the abstraction becomes real. It is the reason a non-engineer trusts what they are buying. It is also the reason the video travels on the internet. We will keep investing in the visual fidelity of the 3D viewer until a stranger watching a 60-second clip believes they understand what they are looking at.

Principle 04
Autonomy is a dial, not a mystery.

Zero is pure manual. One is stabilized. Two is GPS-assisted return-to-home. Three is waypoint mission. Four is on-board decisioning with obstacle avoidance and BVLOS-compliant hardware. Each level changes the BOM, the firmware, and the regulatory envelope. We show the user what they are choosing at every level — what it costs, what it requires, what paperwork it triggers — before they commit.

Principle 05
Drop a photo. Use the part.

Hold up the camera, drag the image in, and we identify the part and slot it into the build. If the new part breaks the old build, we say exactly what breaks and propose what to adjust. If we cannot identify it, we say so and ask. Silence in the face of uncertainty is a failure mode we refuse.

Principle 06
The artifact is portable.

Every design exports to .vdx — a human-readable, archivable, forkable TOML build spec. Your design is yours. If we go out of business, a decade from now, you will still be able to open the file and know what you built.

Principle 07
We iterate small and ship often.

The roadmap is not the product. The product is this week's working thing, used by a real person doing a real job. We would rather ship fixed-wing correctly than promise seven form factors and ship none of them well.

Principle 08
Honest evidence over confident hallucination.

Every part in the catalog carries an evidence level: grounded (verified against a live vendor listing), estimated (derived from family-typical specs), conceptual (rough outline only). The lowest tier wins at the build level and is surfaced to the user. We will never pretend to certainty we do not have.

04How we work

Small. Engineering-heavy. Close to users. We fly the drones we design, we ride along on solar-inspection runs, we talk to farmers and archaeologists and marine biologists. A feature that looks right in Figma but does not survive contact with a real operator is not shipped. We rotate authorship on research publications so the work travels with the product, not with individual CVs. We publish methods because an engineering tool without transparent methods does not deserve the word "engineering."

Our first-year operating principle is that the fastest path to a great product is to be disgustingly attentive to five or six users. If you are one of those users, you will notice. If you are going to be one of them and are not yet, we would love to hear from you.

What we won't build

A design tool is a force multiplier. It is our responsibility to say out loud what we will not multiply.

DRONA will not help design weapons systems, loitering munitions, lethal autonomous systems, or payload dispersal for bio or chemical harm. It will not help design drones whose primary purpose is to evade identification for criminal operations — smuggling, stalking, restricted-airspace intrusion. It will not generate optimized routes or configurations for mass targeting of people, manned or unmanned.

This is not a policy we negotiate with enterprise customers. It is not a feature request we entertain. It is in the system prompt, it is in the tool schemas, and it will remain there.

no weapons systems or lethal payloads no loitering munitions no detection-evasion designs for criminal use no mass-targeting configurations no bio / chemical dispersal systems

We are an engineering tool for people who inspect, film, map, deliver, rescue, scout, and discover. We are happy to keep that line bright and permanent, even if it costs us customers. There are plenty of buyers we are not interested in having.

05Who we're for

We are for the 14-year-old girl on a drone-soccer team. We are for the 52-year-old farmer whose cows keep leaving. We are for the facilities engineer who is tired of paying an outside firm to thermal-scan her own roof. We are for the undergraduate Design-Build-Fly team that wants to stop passing the airframe spec between sophomores and seniors as a PDF. We are for the solar operator in Chile whose storm-triage drone needs to fly in 48 hours with a local operator. We are for the marine biologist who has been told "we have a great quadcopter" one too many times.

We are for the person who has a problem they do not know how to phrase yet. When they finally phrase it in their own words, we want the tool to understand — and to know enough to ask them the two or three questions that would have taken them three weeks on their own to even think of asking.

06Where this goes

The first version is a drone designer. The second version is a problem designer. You describe what you need done. We figure out whether the answer is a drone, a fleet of drones, a tethered platform, an underwater vehicle, a ground rover, or a coordinated combination. Then we spec the combination — honestly, physics-grounded, priced, and buildable.

The third version is a mission network. The farmer who designed the cow-finder can see other farmers running similar rigs, fork their configurations, compare flight logs, and crowd-learn what actually works. This is the GitHub-of-physical-systems pattern, and it is earned, not assumed.

The fourth version is the one that pays for everything else: an enterprise tier for utility inspectors, agritech fleets, and public-safety departments that need audit trails, Blue-UAS-compliant sourcing filters, operator-training records tied to each airframe, and SSO that does not feel like a hostage negotiation.

Each of those is the same product. A chat. A 3D canvas. A compatibility engine. An evidence chain. A portable artifact. The surface area keeps getting more useful without getting more complicated.

07A word about AI

We use language models because they are the best available natural-language interface to structured action. We do not trust them to know physics. We do not trust them to know what part numbers exist. We do not trust them to be honest about their own uncertainty. We trust the catalog, the compatibility engine, the specs engine, and the evidence chain. The AI is the mouth of the system, not its brain. When the AI disagrees with the physics engine, the physics engine wins, every time. This is not a temporary architectural choice; it is the permanent one.

We also do not take AI hype as a substitute for solved problems. Things either work or they do not. A conversation either produces a drone that flies or it does not. The only person who decides that is the user — and the fly test.

08A final note, to ourselves

If we do this well, the tool will disappear behind the work. People will not describe what they used. They will describe what they built. A farmer will say, "I have a drone now that watches the cows." A solar operator will say, "We started doing thermal in-house last year." A university team will say, "Our DBF build file is on GitHub." None of them will mention us by name. That is the measure of success we are working toward.

If we do this badly, we will have added another piece of AI hype to a world that is already drowning in it. We are aware of the risk. We are trying very hard not to do that.

— The DRONA team · April 2026

v1 · revised as we learn

We're hiring for a product that actually matters to us.

Engineering, applied AI, computer graphics, domain experts in agriculture, solar, marine. Small team. Real work.

See open roles → Try the product